Yes that's correct - you can only do distance based restraints with COM groups. You can do regular angle and dihedral restraints just not involving COM groups.
As Feng Pan points out one can use the ncsu_pmd code which was made available as part of update.4 for pmemd.cuda - although this uses different syntax to the nmr restraints. It also uses the upload / download routines on every step so is copying to and from the GPU to the CPU and using the CPU to calculate the restraints on each step. As such the performance may suffer a LOT - although I have not personally checked this.
Note it would be a fairly simple thing I believe to just do the same thing for COM angle and dihedral restraints with the GPU code - just have it upload / download. Could probably add this in a day or so if people are really interested in using it - although this would probably halve performance (maybe more) depending on the size of the system being simulated.
If someone wants to volunteer to write angle and dihedral COM GPU kernels that would be awesome.
All the best
Ross
> On Sep 20, 2016, at 07:53, Carlos Simmerling <carlos.simmerling.gmail.com> wrote:
>
> Did this ever get completed? We want to avoid doing it if someone else has
> already out in the work.
> Thanks
> Carlos
>
> On May 27, 2016 4:30 PM, "Feng Pan" <fpan3.ncsu.edu> wrote:
>
>> Hi, Aasmund Kjendseth
>>
>> The ncsu_pmd can also apply the COMs for angle and dihedral. But it is now
>> only in Sander. I have finished the implementation to PMEMD and PMEMD.cuda.
>> It will be added as patch to Amber16 as soon as in June. Then it will be
>> easy to use the restraints with GPUs.
>>
>> Best
>> Feng Pan
>>
>> On Thu, May 26, 2016 at 8:50 AM, Åsmund Røhr Kjendseth <
>> asmund.rohr.nmbu.no>
>> wrote:
>>
>>> Dear Amber developers,
>>>
>>> in the Amber16 advertisement it was stated “Enhanced NMR restraint
>>> support” for the GPU version. I have been looking forward to use COMs for
>>> angel and dihedral restraints, however, this seems not to be supported
>> yet:
>>>
>>> CUDA (GPU): Implementation does not currently support the use of COM
>>> simulations for angle or dihedral restraints.
>>> Require iat(3) >= 0.
>>>
>>> Could you please indicate whether this is prioritised in future updates
>>> or not?
>>>
>>> Best regards,
>>>
>>> Aasmund Kjendseth
>>>
>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Sep 20 2016 - 08:30:03 PDT
